自组织特征映射网络在模式分类中的应用研究

为了研究自组织特征映射神经网络在对于二维向量进行模式分类时,网络结构的最优化问题,深入研究了SOFM神经网络的结构和算法,说明了SOFM网络的建立方法。以二维向量的模式分类为例,利用所建立的SOFM网络模型对输入的二维向量模式进行分类,研究了输出层节点形状和拓扑结构对分类结果的影响,测试了在不同的训练步数条件下,SOFM模型的权值向量的调整过程和分类效果。仿真结果表明:当网络的输出节点以二维平面形式输出时,长和宽不相等的矩形图的分类性能明显优于正方形图的分类性能,并且在输出节点形式相同的情况下,六边型拓扑结构分类精度明显优于栅格型拓扑结构的SOFM神经网络。

Study of the Application of Self-organizing Feature Mapping Neural Network in Pattern Classification

In this paper the structure and algorithm of SOFM network are discussed in depth to study the question of network structure optimization when SOFM network is applied in pattern classification of two-dimensional vectors. And the establishment of SOFM network is also introduced. The pattern classification of two-dimensional vectors is taken as an example, and their classification is done by SOFM network. The influence of node shape and topology structure of the output layer is under investigation. The adjustment process of weight vectors as well as classification performance of SOFM are also tested in the condition of different number of training steps. The simulation result shows that the classification of rectangles whose length and width are different is better than that of squares, when the output nodes are put out in the form of a two-dimensional plane. And when the output nodes are in the same form, the classification of hexagonal topology is more precise than that of SOFM neural network whose topology is grid-based.